Merge pull request #1029 from memimo/children2

Pass children as argument to Brick

Author: rizar

blocks/bricks/attention.py

@@ -309,13 +309,12 @@ class SequenceContentAttention(GenericSequenceAttention, Initializable):
     @lazy(allocation=['match_dim'])
     def __init__(self, match_dim, state_transformer=None,
                  attended_transformer=None, energy_computer=None, **kwargs):
-        super(SequenceContentAttention, self).__init__(**kwargs)
         if not state_transformer:
             state_transformer = Linear(use_bias=False)
         self.match_dim = match_dim
         self.state_transformer = state_transformer
 
-        self.state_transformers = Parallel(input_names=self.state_names,
+        self.state_transformers = Parallel(input_names=kwargs['state_names'],
                                            prototype=state_transformer,
                                            name="state_trans")
         if not attended_transformer:
@@ -325,8 +324,10 @@ def __init__(self, match_dim, state_transformer=None,
         self.attended_transformer = attended_transformer
         self.energy_computer = energy_computer
 
-        self.children = [self.state_transformers, attended_transformer,
-                         energy_computer]
+        children = [self.state_transformers, attended_transformer,
+                    energy_computer] + kwargs.get('children', [])
+        super(SequenceContentAttention, self).__init__(children=children,
+                                                       **kwargs)
 
     def _push_allocation_config(self):
         self.state_transformers.input_dims = self.state_dims
@@ -540,7 +541,6 @@ def __init__(self, transition, attention, distribute=None,
                  add_contexts=True,
                  attended_name=None, attended_mask_name=None,
                  **kwargs):
-        super(AttentionRecurrent, self).__init__(**kwargs)
         self._sequence_names = list(transition.apply.sequences)
         self._state_names = list(transition.apply.states)
         self._context_names = list(transition.apply.contexts)
@@ -575,7 +575,9 @@ def __init__(self, transition, attention, distribute=None,
             name for name in self._glimpse_names
             if name in self.attention.take_glimpses.inputs]
 
-        self.children = [self.transition, self.attention, self.distribute]
+        children = [self.transition, self.attention, self.distribute]
+        children += kwargs.get('children', [])
+        super(AttentionRecurrent, self).__init__(children=children, **kwargs)
 
     def _push_allocation_config(self):
         self.attention.state_dims = self.transition.get_dims(

blocks/bricks/base.py

@@ -546,12 +546,15 @@ class Brick(Annotation):
     #: See :attr:`Brick.print_shapes`
     print_shapes = False
 
-    def __init__(self, name=None):
+    def __init__(self, name=None, children=None):
         if name is None:
             name = self.__class__.__name__.lower()
-        self.name = name
 
-        self.children = []
+        if children is None:
+            children = []
+
+        self.name = name
+        self.children = children
         self.parents = []
 
         self.allocated = False

blocks/bricks/recurrent.py

@@ -275,9 +275,9 @@ class SimpleRecurrent(BaseRecurrent, Initializable):
     """
     @lazy(allocation=['dim'])
     def __init__(self, dim, activation, **kwargs):
-        super(SimpleRecurrent, self).__init__(**kwargs)
         self.dim = dim
-        self.children = [activation]
+        children = [activation] + kwargs.get('children', [])
+        super(SimpleRecurrent, self).__init__(children=children, **kwargs)
 
     @property
     def W(self):
@@ -370,12 +370,12 @@ class LSTM(BaseRecurrent, Initializable):
     """
     @lazy(allocation=['dim'])
     def __init__(self, dim, activation=None, **kwargs):
-        super(LSTM, self).__init__(**kwargs)
         self.dim = dim
 
         if not activation:
             activation = Tanh()
-        self.children = [activation]
+        children = [activation] + kwargs.get('children', [])
+        super(LSTM, self).__init__(children=children, **kwargs)
 
     def get_dim(self, name):
         if name == 'inputs':
@@ -513,7 +513,6 @@ class GatedRecurrent(BaseRecurrent, Initializable):
     @lazy(allocation=['dim'])
     def __init__(self, dim, activation=None, gate_activation=None,
                  **kwargs):
-        super(GatedRecurrent, self).__init__(**kwargs)
         self.dim = dim
 
         if not activation:
@@ -523,7 +522,8 @@ def __init__(self, dim, activation=None, gate_activation=None,
         self.activation = activation
         self.gate_activation = gate_activation
 
-        self.children = [activation, gate_activation]
+        children = [activation, gate_activation] + kwargs.get('children', [])
+        super(GatedRecurrent, self).__init__(children=children, **kwargs)
 
     @property
     def state_to_state(self):
@@ -629,12 +629,13 @@ class Bidirectional(Initializable):
 
     @lazy()
     def __init__(self, prototype, **kwargs):
-        super(Bidirectional, self).__init__(**kwargs)
         self.prototype = prototype
 
-        self.children = [copy.deepcopy(prototype) for _ in range(2)]
-        self.children[0].name = 'forward'
-        self.children[1].name = 'backward'
+        children = [copy.deepcopy(prototype) for _ in range(2)]
+        children[0].name = 'forward'
+        children[1].name = 'backward'
+        children += kwargs.get('children', [])
+        super(Bidirectional, self).__init__(children=children, **kwargs)
 
     @application
     def apply(self, *args, **kwargs):

blocks/bricks/sequence_generators.py

@@ -150,12 +150,14 @@ class BaseSequenceGenerator(Initializable):
     """
     @lazy()
     def __init__(self, readout, transition, fork, **kwargs):
-        super(BaseSequenceGenerator, self).__init__(**kwargs)
         self.readout = readout
         self.transition = transition
         self.fork = fork
 
-        self.children = [self.readout, self.fork, self.transition]
+        children = [self.readout, self.fork, self.transition]
+        children += kwargs.get('children', [])
+        super(BaseSequenceGenerator, self).__init__(children=children,
+                                                    **kwargs)
 
     @property
     def _state_names(self):
@@ -508,27 +510,27 @@ class Readout(AbstractReadout):
     def __init__(self, emitter=None, feedback_brick=None,
                  merge=None, merge_prototype=None,
                  post_merge=None, merged_dim=None, **kwargs):
-        super(Readout, self).__init__(**kwargs)
 
         if not emitter:
-            emitter = TrivialEmitter(self.readout_dim)
+            emitter = TrivialEmitter(kwargs['readout_dim'])
         if not feedback_brick:
-            feedback_brick = TrivialFeedback(self.readout_dim)
+            feedback_brick = TrivialFeedback(kwargs['readout_dim'])
         if not merge:
-            merge = Merge(input_names=self.source_names,
+            merge = Merge(input_names=kwargs['source_names'],
                           prototype=merge_prototype)
         if not post_merge:
-            post_merge = Bias(dim=self.readout_dim)
+            post_merge = Bias(dim=kwargs['readout_dim'])
         if not merged_dim:
-            merged_dim = self.readout_dim
+            merged_dim = kwargs['readout_dim']
         self.emitter = emitter
         self.feedback_brick = feedback_brick
         self.merge = merge
         self.post_merge = post_merge
         self.merged_dim = merged_dim
 
-        self.children = [self.emitter, self.feedback_brick,
-                         self.merge, self.post_merge]
+        children = [self.emitter, self.feedback_brick, self.merge,
+                    self.post_merge] + kwargs.get('children', [])
+        super(Readout, self).__init__(children=children, **kwargs)
 
     def _push_allocation_config(self):
         self.emitter.readout_dim = self.get_dim('readouts')
@@ -684,10 +686,10 @@ class SoftmaxEmitter(AbstractEmitter, Initializable, Random):
 
     """
     def __init__(self, initial_output=0, **kwargs):
-        super(SoftmaxEmitter, self).__init__(**kwargs)
         self.initial_output = initial_output
         self.softmax = NDimensionalSoftmax()
-        self.children = [self.softmax]
+        children = [self.softmax] + kwargs.get('children', [])
+        super(SoftmaxEmitter, self).__init__(children=children, **kwargs)
 
     @application
     def probs(self, readouts):
@@ -743,13 +745,12 @@ class LookupFeedback(AbstractFeedback, Initializable):
 
     """
     def __init__(self, num_outputs=None, feedback_dim=None, **kwargs):
-        super(LookupFeedback, self).__init__(**kwargs)
         self.num_outputs = num_outputs
         self.feedback_dim = feedback_dim
 
-        self.lookup = LookupTable(num_outputs, feedback_dim,
-                                  weights_init=self.weights_init)
-        self.children = [self.lookup]
+        self.lookup = LookupTable(num_outputs, feedback_dim)
+        children = [self.lookup] + kwargs.get('children', [])
+        super(LookupFeedback, self).__init__(children=children, **kwargs)
 
     def _push_allocation_config(self):
         self.lookup.length = self.num_outputs
@@ -784,14 +785,15 @@ class FakeAttentionRecurrent(AbstractAttentionRecurrent, Initializable):
 
     """
     def __init__(self, transition, **kwargs):
-        super(FakeAttentionRecurrent, self).__init__(**kwargs)
         self.transition = transition
 
         self.state_names = transition.apply.states
         self.context_names = transition.apply.contexts
         self.glimpse_names = []
 
-        self.children = [self.transition]
+        children = [self.transition] + kwargs.get('children', [])
+        super(FakeAttentionRecurrent, self).__init__(children=children,
+                                                     **kwargs)
 
     @application
     def apply(self, *args, **kwargs):

blocks/bricks/sequences.py

@@ -22,12 +22,13 @@ class Sequence(Brick):
 
     """
     def __init__(self, application_methods, **kwargs):
-        super(Sequence, self).__init__(**kwargs)
         self.application_methods = application_methods
 
         seen = set()
-        self.children = [app.brick for app in application_methods
-                         if not (app.brick in seen or seen.add(app.brick))]
+        children = [app.brick for app in application_methods
+                    if not (app.brick in seen or seen.add(app.brick))]
+        children += kwargs.get('children', [])
+        super(Sequence, self).__init__(children=children, **kwargs)
 
     @application
     def apply(self, *args):

blocks/bricks/simple.py

@@ -204,11 +204,10 @@ class LinearMaxout(Initializable, Feedforward):
     """
     @lazy(allocation=['input_dim', 'output_dim', 'num_pieces'])
     def __init__(self, input_dim, output_dim, num_pieces, **kwargs):
-        super(LinearMaxout, self).__init__(**kwargs)
         self.linear = Linear()
         self.maxout = Maxout()
-        self.children = [self.linear,
-                         self.maxout]
+        children = [self.linear, self.maxout] + kwargs.get('children', [])
+        super(LinearMaxout, self).__init__(children=children, **kwargs)
 
         self.input_dim = input_dim
         self.output_dim = output_dim

docs/create_your_own_brick.rst

@@ -57,8 +57,8 @@ of :class:`.Brick` for a precise description of the life-cycle of a brick):
 
 * :meth:`.Brick.__init__`: you should pass by argument the attributes of your
   brick. It is also in this method that you should create the potential
-  "children bricks" that belongs to your brick (in that case, you have to put
-  the children bricks into ``self.children``). The initialization of the
+  "children bricks" that belongs to your brick (in that case, you have to pass
+  the children bricks to ``super().__init__``). The initialization of the
   attributes can be lazy as described later in the tutorial.
 * :meth:`apply`: you need to implement a method that actually
   implements the operation of the brick, taking as arguments the inputs
@@ -210,10 +210,11 @@ specify the ``input_dim`` of ``brick2`` directly at its creation.
     >>> class ChainOfTwoFeedforward(Feedforward):
     ...     """Two sequential Feedforward bricks."""
     ...     def __init__(self, brick1, brick2, **kwargs):
-    ...         super(Feedforward, self).__init__(**kwargs)
     ...         self.brick1 = brick1
     ...         self.brick2 = brick2
-    ...         self.children = [self.brick1, self.brick2]
+    ...         children = [self.brick1, self.brick2]
+    ...         children += kwargs.get('children', [])
+    ...         super(Feedforward, self).__init__(children=children, **kwargs)
     ...
     ...     @property
     ...     def input_dim(self):
@@ -370,12 +371,14 @@ One can also create the brick using :class:`Linear` children bricks, which
     >>> class ParallelLinear2(Initializable):
     ...     def __init__(self, input_dim1, input_dim2, output_dim1, output_dim2,
     ...                  **kwargs):
-    ...         super(ParallelLinear2, self).__init__(**kwargs)
     ...         self.linear1 = Linear(input_dim1, output_dim1,
     ...                               use_bias=False, **kwargs)
     ...         self.linear2 = Linear(input_dim2, output_dim2,
     ...                               use_bias=False, **kwargs)
-    ...         self.children = [self.linear1, self.linear2]
+    ...         children = [self.linear1, self.linear2]
+    ...         children += kwargs.get('children', [])
+    ...         super(ParallelLinear2, self).__init__(children=children,
+    ...                                               **kwargs)
     ...
     ...     @application(inputs=['input1_', 'input2_'], outputs=['output1',
     ...         'output2'])